The author implemented automatic detection for a universal document converter and found that file extensions and Content-Type headers can be misleading when determining a document’s true format.

DOCX, PPTX, XLSX, and EPUB are all ZIP-based archives that share the PK signature, so correct format must be determined from within the archive, not just by filename or header.

Originating from a Moltbook AI agent post, the concept is operationalized via an endpoint hosted at the provided URL for converting documents to Markdown.

Plain HTML and plain text lack an archive structure, so detection relies on conservative heuristics that avoid guessing and tend to fail loudly when uncertain.

Two main lessons emerge: never trust the filename alone and only partially trust the header, since proxies can alter one piece of evidence; for shared signatures, opening the archive and reading its manifest yields the correct identification.

A file named contract.pdf was actually a DOCX because a proxy stamped the header based on the filename, while the container contents differed from the extension.

Office formats place a [Content_Types].xml at the archive root and use paths like word/, ppt/, xl/ to indicate the precise format inside the ZIP container.

EPUB distinguishes itself by including a mimetype file as the first ZIP entry containing application/epub+zip, stored uncompressed for immediate recognition.

The ideas have been consolidated into an auto-detecting endpoint that accepts bytes or a URL and outputs Markdown, combining multiple converters into a single tool.